[PATCH] Revert "drm/amdgpu: don't modify num_doorbells for mes"
Sharma, Shashank
Shashank.Sharma at amd.com
Fri Aug 4 07:36:50 UTC 2023
[AMD Official Use Only - General]
Nack to the revert.
This is a whole series of doorbell changes, which has replacement for the patches too.
https://patchwork.freedesktop.org/series/115802/
The whole series was pushed to staging branch, but looks like some of the patches did not merge.
When the whole series is merged, functionality will be restored.
I will check if there is any problem due to which other patches are blocked merge.
Regards
Shashank
-----Original Message-----
From: Zhang, Yifan <Yifan1.Zhang at amd.com>
Sent: Friday, August 4, 2023 7:01 AM
To: amd-gfx at lists.freedesktop.org
Cc: Deucher, Alexander <Alexander.Deucher at amd.com>; Koenig, Christian <Christian.Koenig at amd.com>; Sharma, Shashank <Shashank.Sharma at amd.com>; Zhang, Yifan <Yifan1.Zhang at amd.com>
Subject: [PATCH] Revert "drm/amdgpu: don't modify num_doorbells for mes"
This reverts commit f46644aa8de6d5efeff8d8c7fbf3ed58a89c765c.
THe doorbell index could go beyond the first page for mes queues, this patch breaks the mes self test on gfx11.
[ 23.212740] [drm] ring gfx_32768.1.1 was added
[ 23.213147] [drm] ring compute_32768.2.2 was added
[ 23.213540] [drm] ring sdma_32768.3.3 was added
[ 23.213546] [drm:amdgpu_mm_wdoorbell64 [amdgpu]] *ERROR* writing beyond doorbell aperture: 0x00001000!
[ 23.214148] amdgpu 0000:c2:00.0: amdgpu: gfx_v11_0_ring_set_wptr_gfx: 5168 0x402 0x1000 100
[ 23.560357] amdgpu 0000:c2:00.0: [drm:amdgpu_ring_test_helper [amdgpu]] *ERROR* ring gfx_32768.1.1 test failed (-110)
Signed-off-by: Yifan Zhang <yifan1.zhang at amd.com>
---
.../gpu/drm/amd/amdgpu/amdgpu_doorbell_mgr.c | 34 +++++++++++--------
1 file changed, 19 insertions(+), 15 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell_mgr.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell_mgr.c
index 5c0d3cea817d..31db526d4921 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell_mgr.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell_mgr.c
@@ -140,21 +140,25 @@ int amdgpu_doorbell_init(struct amdgpu_device *adev)
adev->doorbell.base = pci_resource_start(adev->pdev, 2);
adev->doorbell.size = pci_resource_len(adev->pdev, 2);
- adev->doorbell.num_kernel_doorbells =
- min_t(u32, adev->doorbell.size / sizeof(u32),
- adev->doorbell_index.max_assignment + 1);
- if (adev->doorbell.num_kernel_doorbells == 0)
- return -EINVAL;
-
- /*
- * For Vega, reserve and map two pages on doorbell BAR since SDMA
- * paging queue doorbell use the second page. The
- * AMDGPU_DOORBELL64_MAX_ASSIGNMENT definition assumes all the
- * doorbells are in the first page. So with paging queue enabled,
- * the max num_kernel_doorbells should + 1 page (0x400 in dword)
- */
- if (adev->asic_type >= CHIP_VEGA10)
- adev->doorbell.num_kernel_doorbells += 0x400;
+ if (adev->enable_mes) {
+ adev->doorbell.num_kernel_doorbells =
+ adev->doorbell.size / sizeof(u32);
+ } else {
+ adev->doorbell.num_kernel_doorbells =
+ min_t(u32, adev->doorbell.size / sizeof(u32),
+ adev->doorbell_index.max_assignment+1);
+ if (adev->doorbell.num_kernel_doorbells == 0)
+ return -EINVAL;
+
+ /* For Vega, reserve and map two pages on doorbell BAR since SDMA
+ * paging queue doorbell use the second page. The
+ * AMDGPU_DOORBELL64_MAX_ASSIGNMENT definition assumes all the
+ * doorbells are in the first page. So with paging queue enabled,
+ * the max num_kernel_doorbells should + 1 page (0x400 in dword)
+ */
+ if (adev->asic_type >= CHIP_VEGA10)
+ adev->doorbell.num_kernel_doorbells += 0x400;
+ }
adev->doorbell.ptr = ioremap(adev->doorbell.base,
adev->doorbell.num_kernel_doorbells *
--
2.37.3
More information about the amd-gfx
mailing list